Accurate photovoltaic (PV) power forecasting is crucial for the efficient utilization of solar energy and the provision of low-carbon power in electrified railways. To improve prediction accuracy and reduce lag caused by the stochastic fluctuations of railway-side PV systems, this paper proposes a hybrid GWO-VMD-CNN-BiGRU-Attention model. First, the Grey Wolf Optimizer (GWO) optimizes the parameters of Variational Mode Decomposition (VMD), which adaptively decomposes PV output into stable sub-modal components based on fuzzy entropy (FE). Each component is then individually forecasted using a CNN-BiGRU-Attention network: the CNN extracts temporal features, the BiGRU captures dynamic patterns, and the attention mechanism highlights critical time steps. The final prediction is obtained by summing the component forecasts. Validated on real-world data from a high-speed railway, the model effectively mitigates prediction lag and outperforms benchmark methods in accuracy.

Grey Wolf Optimizer (GWO), Hybrid model forecasting, Photovoltaic power forecasting ,Variational Mode Decomposition (VMD)